90% Token Cost Cut Needed for AI Startups to Survive, Says Palo Alto CEO
Key Takeaways
- For AI startups, unsustainable token costs are burning runway and stalling growth.
- Nikesh Arora’s call for a 90% reduction signals a potential market shift that could determine which startups survive to scale.
Mentioned
Key Intelligence
Key Facts
- 1Nikesh Arora stated token costs need to drop 20% over the next 12 months and 90% by the following year for large-scale enterprise AI adoption (CNBC, July 9, 2026).
- 2Sam Altman claimed OpenAI's latest model achieves 54% greater token efficiency for agentic coding, which Arora called "a good start" but insufficient to meet enterprise needs.
- 3Uber exhausted its full-year 2026 AI budget by April, forcing a review of token costs versus human engineer hiring, exemplifying enterprise budget strain.
- 4Palantir CEO Alex Karp criticized the token model of Anthropic and OpenAI as "completely wrong" and advocated for open-weight alternatives, echoing Arora's concerns (CNBC, July 2, 2026).
- 5Chinese AI labs are gaining traction with cheaper models due to lower energy costs and greater efficiency, presenting a cost advantage over U.S. providers.
- 6Tech giants raised massive debt to fund AI infrastructure: SpaceX completed a $25 billion bond sale and Amazon a $25 billion debt offering in mid-2026.
Who's Affected
Arora's target to unlock mass enterprise AI adoption and startup scalability
Analysis
Founders and VCs have been grappling with a silent runway killer: the soaring cost of AI tokens. Now, a prominent voice from the enterprise world is putting numbers on the crisis. Nikesh Arora, CEO of Palo Alto Networks, says token costs need to fall 90% for broad enterprise AI adoption—a metric that could make or break the next generation of AI startups. For early-stage companies, the clock is ticking: can they survive the current burn rate long enough to see the predicted cost relief?
In a stark warning issued during a CNBC interview on July 9, 2026, Palo Alto Networks CEO Nikesh Arora declared that token costs—the price enterprises pay for each unit of AI model usage—need to plummet by 90% within two years to enable large-scale AI adoption. His comments, which came on the heels of OpenAI CEO Sam Altman’s claim that the company’s newest model is 54% more efficient for coding tasks, underscored a growing anxiety among business leaders that the current economics of AI are unsustainable. Arora’s call for a 20% reduction in the next 12 months and a 90% cut by the following year reflects a reckoning that the AI industry’s pricing model is at odds with enterprise scalability.
SpaceX’s $25 billion bond sale and Amazon’s $25 billion debt raise this month are emblematic of the financial strain.
The backdrop is a surge in AI spending that has caught many companies off guard. As reported in May 2026, “token shock” hit major users like Uber, which exhausted its entire 2026 AI budget by April. Uber’s CTO described going “back to the drawing board,” and the COO said the company would now weigh token costs directly against hiring human engineers. This anecdote is a microcosm of a broader trend: enterprises that initially encouraged unfettered AI experimentation are now implementing usage caps, switching to older, cheaper models, and exploring open-source alternatives. The PYMNTS report from June noted that companies are urgently seeking ways to manage costs, including looking to Chinese AI labs that offer lower prices due to more efficient models and China’s cheaper energy.
Arora is not alone. Palantir CEO Alex Karp, in a separate CNBC interview just a week earlier, lambasted the token-based pricing model of frontier labs like Anthropic and OpenAI, saying “something has gone completely wrong” and accusing them of wasting enterprises’ time. Karp championed open-weight models as a solution, a sentiment that Arora implicitly supports given his push for dramatic cost declines. The collective frustration from these tech leaders signals that the AI market is approaching a tipping point where pricing must become more aligned with the value delivered, or enterprises will defect to alternatives.
What to Watch
The implications are profound. For AI adoption to move from experimental projects to widespread integration, token costs must fall steeply. This will pressure model providers to improve efficiency—as Altman’s 54% improvement shows—but Arora believes that’s not nearly enough. The competitive landscape is shifting: Chinese labs are closing the gap, and their cost advantage could erode the dominance of U.S.-based providers if pricing remains high. Meanwhile, the massive capital expenditures by tech giants to build AI infrastructure are creating a paradox: soaring investment in capacity alongside a customer base that is balking at the price of using it. SpaceX’s $25 billion bond sale and Amazon’s $25 billion debt raise this month are emblematic of the financial strain.
Looking forward, Arora predicts that either the market will force a correction in spending or businesses will adjust. Efficiency improvements and competitive pressures will likely drive token costs down over time, but the pace may be slower than enterprises need. The emergence of open-source models and Chinese alternatives could accelerate this trend, though they introduce their own risks around quality, security, and geopolitical concerns. Ultimately, the next two years will be critical in determining whether AI becomes a ubiquitous enterprise tool or remains a costly luxury for the few with deep pockets. For cybersecurity—Palo Alto’s own domain—AI-driven threat detection and response are becoming essential, but token costs could stall deployments just as attack surfaces expand. For startups, the high cost of AI is a direct threat to runway and valuation. For the broader AI industry, the call for a 90% reduction sets a new benchmark that will likely reshape product roadmaps and pricing strategies. The token cost debate is no longer a niche concern; it is a defining issue for the next phase of enterprise technology.
From the Network
How we covered this story
Every story in our startup coverage is assembled from multiple primary sources, cross-referenced for factual consistency, and scored along three independent dimensions: sentiment, operational impact, and source-cluster confidence. Single-source rumors and unverifiable claims do not pass our editorial gate. When a story shows "Verified by N sources" with N≥2, the development is independently corroborated; when N=1, we mark it explicitly so readers can weigh the signal accordingly.
Impact scoring uses a 1-10 scale weighted toward regulatory, financial, and operational consequence rather than coverage volume. A topic that runs in every outlet but moves no real decisions ranks lower than a niche regulatory filing that reshapes how operators in the startup space have to behave. Read our full methodology for the scoring rubric, our glossary for term definitions, and our trends index for the longitudinal view across the beat.
| Signal on this page | What it tells you |
|---|---|
| Verified by N sources | Independent corroboration count. N≥2 is our confidence floor; N=1 is marked explicitly. |
| Impact score (1-10) | Regulatory + financial + operational weight. 8+ signals an experienced-operator action item. |
| Sentiment | Five-tier classification trained on labeled startup-specific corpora. |
| Timeline | Where applicable, the related-events sequence that contextualizes today's development. |